Clustering techniques for object-oriented database systems

Abstract

In an Object-Oriented Database system (OODB), the interrelationships among objects and inheritance semantics can be used by clustering techniques to improve the performance in terms of system response time. In this paper, we conduct an analysis to compare three clustering strategies, Cactis, ORION, and CK in terms of space and time overhead. We also examine the level, page or segment, at which clustering should take place. The dynamic clustering technique CK is found to be best in exploiting the structural relationships between objects and inheritance semantics to identifi an efficient storage scheme. However it creates high overhead and is only best suited for applications in which the Readprite ratio is high. To remove this limitation, we present how segment clustering instead of page clustering could reduce the number of cases where a page split is necessa y.

A comparison study of object-oriented database clustering techniques

“…This is an advantage since no arbitrary choice has to be made by the user [CHAB93]. But it also implies some time overhead (time to compute total number of times each object is accessed and number of times each relationship is crossed) and space overhead (the main memory space used to store the counters grows with the database size).…”

“…But simplicity also turns to a limitation since users' hints can only be based on the static information given by the data model and not on some information determined by the database usage and which could lead to a better clustering [CHAB93]. Figure 4 illustrates the way we implemented the ORION clustering method in our simulation models.…”

“…When a segment is full, a new page is allocated and linked to the segment (a pointer must be maintained in the segment descriptor). This implies some overhead to find the address of each additional page [CHAB93].…”

“…This implies that the database is not clustered on the first run because no information about the database is available [CHAB93].…”

“…However, in OODBs, this form of inheritance (called type inheritance) is not sufficient. [CHAB93] The CK algorithm also uses instance-to-instance inheritance that not only transfers the existence of attributes from one object to another (like type inheritance), but moreover the values of these attributes [WILK88].…”